Vicom Infinity
Voice Assistant
Digital voice assistants are transforming the way people receive information, control their lives, and communicate, whether they're on a mobile phone or a home-based gadget. When we look at the market, we see that all businesses are using voice-first technologies to improve customer experience and interaction.
The Challenge
Secure, custom, helpful
We started from the ground up with the goal of providing the best interaction experience and nailing conversational design. We were tasked with determining the most efficient method for extracting data from devices in a corporate context while keeping all sensitive data offsite.
The Idea
Meet VIVA: The Vicom Infinity Voice Assistant
Always at your service
How do you go about doing this? By using your own voice!
Handle sensitive data offline and securely, enable users to add their own actions, and give them full control over the endpoint the device hits to find the right answers.
A level of customization that Alexa or Google will never provide.
The Solution
Train a custom-built AI that recognizes when you call it, fully offline for optimal performance and security. Only when triggered, connect to a backend to process questions and speak the answer. Run it on a mobile phone, as well as a custom-built hardware unit.
Our approach
We focused on:
- Providing maximum value to our clients by conducting thorough market research and developing the appropriate user personas (including sample dialogs).
- Delivering a natural experience - we were able to make interactions as seamless as possible by doing comprehensive user testing. Proper use of Natural language processing (NLP) allowed us to create a genuinely conversational experience that provides shortcuts to the most relevant use cases.
Custom from start to finish
PocketSphinx, Snowboy, and TensorFlow were the three methods we had to try. We ended up building our own Tensorflow model because both PocketSphinx and Snowboy were significantly below our product quality standard. We put each model to the test, and Tensorflow came out on top with a score of 98.3 %. Now that's something we can ship!
To determine the best alternative for natural language processing, we established two voice agents, one on Google's DialogFlow and the other on IBM's Watson. It took a lot of testing, but DialogFlow outperformed the competition, so it was an easy pick.
UTRECHT
+31 30 800 04 12
welcome@pixplicity.com
Molensteyn 60B, 3rd floor
3454PT, Utrecht
The Netherlands